key: cord-0966791-m13ndhaz
authors: Burnett, Deborah L.; Jackson, Katherine J.L.; Langley, David B.; Aggrawal, Anupria; Stella, Alberto Ospina; Johansen, Matt D.; Balachandran, Harikrishnan; Lenthall, Helen; Rouet, Romain; Walker, Gregory; Saunders, Bernadette M.; Singh, Mandeep; Li, Hui; Henry, Jake Y.; Jackson, Jennifer; Stewart, Alastair G.; Witthauer, Franka; Spence, Matthew A.; Hansbro, Nicole G.; Jackson, Colin; Schofield, Peter; Milthorpe, Claire; Martinello, Marianne; Schulz, Sebastian R.; Roth, Edith; Kelleher, Anthony; Emery, Sean; Britton, Warwick J.; Rawlinson, William D.; Karl, Rudolfo; Schäfer, Simon; Winkler, Thomas H.; Brink, Robert; Bull, Rowena A.; Hansbro, Philip M.; Jäck, Hans-Martin; Turville, Stuart; Christ, Daniel; Goodnow, Christopher C.
title: Immunizations with diverse sarbecovirus receptor binding domains elicit SARS-CoV-2 neutralizing antibodies against a conserved site of vulnerability
date: 2021-10-29
journal: Immunity
DOI: 10.1016/j.immuni.2021.10.019
sha: f6c3b88e08fb43218896f3834ba172c695623221
doc_id: 966791
cord_uid: m13ndhaz

Viral mutations are an emerging concern in reducing SARS-CoV-2 vaccination efficacy. Second generation vaccines will need to elicit neutralizing antibodies against sites that are evolutionarily conserved across the sarbecovirus subgenus. Here, we immunized mice containing a human antibody repertoire with diverse sarbecovirus receptor binding domains (RBDs) to identify antibodies targeting conserved sites of vulnerability. Antibodies with broad reactivity against diverse clade B RBDs targeting the conserved class 4 epitope, with recurring IGHV/IGKV pairs, were readily elicited but were non-neutralizing. However, rare class 4 antibodies binding this conserved RBD supersite showed potent neutralization of SARS-CoV-2 and all variants of concern. Structural analysis revealed that neutralizing ability of cross-reactive antibodies was reserved only for those with an elongated CDRH3 that extends the antiparallel beta-sheet RBD core and orients the antibody light chain to obstruct ACE2-RBD interactions. These results identify a structurally defined pathway for vaccine strategies eliciting escape-resistant SARS-CoV-2 neutralizing antibodies.

* Lead contact d.burnett@garvan.org.au. †= Joint senior authors

Viral mutations are an emerging concern in reducing SARS-CoV-2 vaccination efficacy.

Second generation vaccines will need to elicit neutralizing antibodies against sites that are evolutionarily conserved across the sarbecovirus subgenus. Here, we immunized mice containing a human antibody repertoire with diverse sarbecovirus receptor binding domains (RBDs) to identify antibodies targeting conserved sites of vulnerability. Antibodies with broad reactivity against diverse clade B RBDs targeting the conserved class 4 epitope, with recurring IGHV/IGKV pairs, were readily elicited but were non-neutralizing. However, rare class 4 antibodies binding this conserved RBD supersite showed potent neutralization of SARS-CoV-2 and all variants of concern. Structural analysis revealed that neutralizing ability of cross-reactive antibodies was reserved only for those with an elongated CDRH3 that extends the antiparallel beta-sheet RBD core and orients the antibody light chain to obstruct

The current generation of COVID-19 vaccines provide strong protection against severe disease, but the lack of broad cross-neutralization elicited by the vaccine is a source of concern, as future variants are likely to emerge. Only a small percentage of antibodies against the spike protein on the SARS-CoV-2 envelope are virus-neutralizing, the majority of which are directed against the receptor binding domain (RBD) . Among these, class 1 and 2 antibodies predominate, focused against epitopes in the ACE2 receptor binding site (RBS) (Barnes et al., 2020) . The RBS is highly variable while preserving ACE2 binding, and only 48% conserved between SARS-CoV-2 (abbreviated here as CoV2) and SARS-CoV-1 (CoV1) compared to 84% amino acid conservation for non-RBS regions of the RBD . K417N/T, E484K, and N501Y mutations in the RBS epitope have independently arisen in rapidly spreading SARS-CoV-2 variants-of-concern in South Africa (B.1.351/beta) and Brazil/Japan (P.1/gamma) Starr et al., 2021) . These mutations decrease the serum neutralization titer of mRNA-vaccinated people by up to 70% (Garcia-Beltran et al., 2021; Ikegame et al., 2021; Liu et al., 2021; Wang et al., 2021; . This has been shown to have clinical impacts limiting vaccine efficacy (Hacisuleyman et al., 2021; Madhi et al., 2021) . The ongoing risk of new zoonotic spillover events from sarbecoviruses circulating widely in bats also remains a major concern (Banerjee et al., 2021) .

In order to address this problem, second generation vaccines will need to elicit neutralizing antibodies directed against sites that are evolutionarily conserved across the sarbecovirus subgenus (clade B) including CoV2, the related bat virus RaTG13, and the more distant CoV1, which differ at key residues characterizing variants of concern such as K417, E484 and N501 (Boni et al., 2020) (TableS1B) . Here, we analyzed broadly reactive antibody responses elicited by RBD-focused immunization of mice that carry human antibody gene segments (Asensio et al., 2019; Peter et al., 2021) , employing multi-color flow-cytometric staining of B cells with RBDs from CoV1, CoV2, and RaTG13, coupled with single-cell RNA sequencing and antibody characterization. We found that for immunization strategies utilizing diverse sarbecovirus RBDs, class 4 antibodies dominated the broadly reactive subset and used recurring human IGHV and IGKV elements. However, such antibodies were generally not neutralizing, as previously observed for CR3022 (ter Meulen et al., 2006) . In contrast, potent neutralizing antibodies binding the highly conserved class 4 epitope that J o u r n a l P r e -p r o o f sterically blocks the RBS from accessing ACE2, employing a long CDRH3 and less common IGHV and IGKV pairs, could be expanded by multiple immunizations with CoV1 RBD.

These results provide a guide for developing second generation COVID-19 vaccine strategies.

Immunization with diverse sarbecovirus antigens induced a cross-reactive response in wild type mice

The current CoV2 vaccines utilize the whole spike protein to induce clonal selection of antibody-forming B cells. Given the goal of eliciting antibodies specifically to highly conserved regions of the RBD, we explored the capacity of isolated CoV2 RBD compared to full trimeric spike, enhancing immunogenicity by conjugation to sheep red blood cells (SRBCs) (FigS1) . Previous experience has shown that small foreign proteins elicit more reproducible germinal center (GC) B cell populations when covalently linked to SRBCs without adjuvant than when given in adjuvant as isolated proteins or covalently linked to other immunogenic carrier proteins. Mice from C57BL/6, BALB/c and FVB/NJ strains with distinct Ighv and Igkv loci (Collins et al., 2015) were immunized with equivalent density of conjugated CoV2 RBD or spike (FigS1A) and spleen cells analyzed 7 days later by staining with CoV2 RBD fluorescent tetramers, which selectively bound to individual B cells bearing surface immunoglobulin (Ig) with affinity for CoV2 RBD epitopes. Immunization of mice with RBD or spike resulted in the development of RBD specific B cells in the GC, IgG1 memory and plasmablast compartment (FigS1B). Full trimeric spike resulted in greater recruitment of RBD specific cells in the GC compartment, while both antigens were equally efficient at recruiting these cells into the memory and plasmablast compartment.

To enumerate B cells with broadly reactive surface Ig within the elicited RBD response, these analyses were extended by flow cytometric staining with a panel of distinguishable fluorescent RBD tetramers from CoV2, Pangolin, RaTG13 and CoV1, representing progressively more distant RBD sequences (tableS1A). 50-70% of COV2 RBD-binding B cells in the CoV2-RBD-elicited GC and memory repertoires were cross-reactive to pangolin derived RBD, 30-60% were cross-reactive to RaTG13, but only 5-15% displayed crossreactivity to CoV1 (Fig1A, B, FigS2A) . For comparison, the same flow cytometric analysis was performed following intranasal infection with 1 x 10 4 PFU SARS-CoV-2 in K18-hACE2-C57BL/6 mice with humanized ACE2 receptors (Johansen et al., 2020) . This revealed analogous results with 60% of CoV2 RBD-binding GC B cells cross-reacting to pangolin RBD, 19% cross-reacting RaTG12 and 1.8% to SARS-CoV-1 (FigS1B,C).

Given that CoV2 RBD immunization elicited cross-reactive GC and memory B cells binding to other sarbecovirus RBDs, we next tested if the inverse also applied (Fig1C). Immunization J o u r n a l P r e -p r o o f with RaTG13 RBD elicited a strong CoV2 RBD binding response. Immunization with the more distant CoV1 RBD elicited RBD-binding B cells that were mostly specific to the CoV1 RBD, however a clear subset of the CoV1-binding cells cross-reacted with all three sarbecovirus RBDs. All three immunizations induced a subset of B cells with surface Ig that is triple-reactive: binding CoV2, RaTG13, and CoV1 RBDs (Fig1D) .

Overall, we found that both spike and RBD induced a CoV2 specific response. A proportion of the response following CoV2 RBD immunization or infection included B cells with broad cross-reactivity. These broadly reactive B cells were also induced by immunization with CoV1 or RaTG13 RBDs.

Immunization with diverse sarbecovirus RBDs induced a cross-reactive response in human antibody VDJ transgenic mice We next extended this analysis to C57BL/6 TRIANNI transgenic mice where the human antibody variable, diversity and joining (VDJ) element repertoire replaces the mouse VDJ repertoire at the heavy chain and kappa light chain loci (Asensio et al., 2019; Peter et al., 2021) . Immunization of the Ig-humanized mice with CoV2 RBD conjugated to SRBCs elicited a GC response with 79% of CoV2-binding GC B cells cross-reactive to the closely related pangolin RBD, 42% to RaTG13 and 10% to the more distant CoV1 RBD (Fig1F).

Again, 0.005-0.015% of all GC B cells displayed antibodies that cross-reacted to all 3 RBDs (Fig1H). To prime-boost the response, the Ig-humanized mice were immunized with RBD conjugated to SRBCs on day 0 and 6 and then with the same RBD-conjugated to a different adjuvant-free immunogenic carrier, horse red blood cells (HRBCs), on days 10 and 15, and the effects explored on day 20 (Fig1E-H). Prime-boosting increased by 10-fold the percentage of GC B cells binding CoV2 RBD (Fig1G) and cross-reacting with CoV2, RaTG13 and CoV1 (Fig1H) .

We extended the flow cytometric approach to analyze which RBD epitope is recognized by each B cell responding to the different sarbecovirus RBD immunizations. Spleen cells were first incubated with equimolar monomeric CoV2, CoV1 or RaTG13 RBD to identify B cells with surface Ig capable of binding each antigen. Bound RBD was revealed by staining the cells with fluorescent S309 binding the class 3 epitope (Pinto et al., 2020) , fluorescent EY6A recognizing the class 4 epitope (Zhou et al., 2020) , and fluorescent ACE2 (Fig2) TableS2A-C) and from an additional set of 52 triple RBD tetramer-binding B cells sorted from separate mice (FigS4A, TableS2D). Across these immunization strategies several IGHV/IGKV pairs were frequently used by GC B cells that cross-reacted with all three sarbecovirus RBDs: VH1-18 paired with VK6-21 or VK6D-21, VH1-46 paired with VK1-9 or VK1-6, and VH3-33 paired with VK1D-13 or FigS4A) .

For comparison with the RBD-elicited B cells, we obtained paired IGHV/IGKV sequences from 6000 naïve B cells from the humanized mice and analyzed deep cDNA sequencing of the expressed IgM repertoires of 10 6 cells from two additional humanized mice and in naïve B cells from 100 human blood samples (FigS4C). This confirmed Ig-humanized mice express diverse IGHVs. Although some IGHV elements like IGHV1-69 showed an altered frequency compared to humans, 78% of IGHV elements were used at comparable frequency to the naïve repertoire in human blood, including many of those frequently used in RBD binding antibodies: IGHV1-18, . IGHV3-53, which accounts for frequent germline-encoded antibodies against the hypervariable RBS region of CoV2 (Andreano and Rappuoli, 2021; Yuan et al., 2020) , was strongly selected among the total set of sequenced RBD-binding B cells when compared to its frequency in the naïve repertoire, but was rarely used among the highly cross-reactive triple-RBD binders from the humanized mice (Fig3B-E, Fig S4A) . The latter result is consistent with RBD-binding antibodies isolated from SARS-CoV-2 infected people and deposited in CoV-AbDab (Raybould et al., 2021) , where IGHV3-53 is frequent among CoV2specific antibodies but rarely used among antibodies that also bind CoV1 (FigS4D).

For comparison to the Ig-humanized mice, rare triple RBD tetramer-binding memory B cells were sorted from the blood of SARS-CoV-2 convalescent human patients (Fig4A,B) . Single cell sequencing of 121 CoV2, CoV1 and RaTG13 triple-binders revealed that many used IGHV regions observed in triple binders from RBD-immunized Ig-humanized mice, including the recurring IGHV1-46, , Table S2H ). Indeed, 92% of the IGHVs identified from the 53 triple binding cells sorted from Ig-humanized mice were utilized among the 121 human B cells sorted with the same strategy. These IGHV elements are also used by CoV1/CoV2 RBD cross-reactive antibodies isolated from SARS-CoV-2 infected people and deposited in CoV-AbDab, confirming cross-reactive B cells selected in the humanized mouse repertoire to be representative of the human repertoire (FigS4D).

From the 7533 RBD-binding antibody sequences obtained from RBD-immunized Ighumanized mice, a total of 56 antibodies were selected for expression as human IgG1, representing IGHV/IGKV pairs found in >1% of sarbecovirus RBD-binding B cells or pairs recurrently selected between different immunization regimes (TableS3A). Antibodies were tested for binding to different sarbecovirus RBDs by flow cytometry against RBD-conjugated erythrocytes (Fig5A,B, TableS2A). Previously described class 4 antibodies, CR3022 (ter Meulen et al., 2006) and EY6A (Zhou et al., 2020) , and the class 3 antibody S309 (Pinto et al., 2020) were expressed and tested in parallel as controls. Overall, 15 antibodies bound all three RBDs (Fig5A,B) . Most used the recurring IGHV/IGKV pairs noted above except AB-3467, which used VH4-59 paired with VK1-9 and was part of a very large, heavily mutated clonal expansion of cells binding CoV1 and CoV2 RBDs that comprised 7.9% of the RBDbinding GC cells sorted from a CoV1 RBD immunized mouse (Fig5C,D). Of the IGHV regions that accounted for recurrently selected class 4 sequences IGHV1-18, , which together accounted for 76% of class 4 antibodies, were equally represented in the human and mouse naïve J o u r n a l P r e -p r o o f repertoires. IGHV3-33 which accounted for 16% of the class 4 sequences was overrepresented in the mouse naïve repertoire. IGHV4-59 which accounted for 8% of class 4 antibodies and AB-3467, was underrepresented in the mouse naïve repertoire (FigS4D).

Together, this revealed that antibodies cross-reactive to multiple RBDs predominantly used a subset of recurrently selected IGHV/IGKV pairs. The total 36 antibodies binding strongly to CoV2 were tested at four concentrations for neutralization of lentivirus particles pseudotyped with CoV2 spike, alongside EY6A, S309 and CR3022. Only four were potent neutralizers (half maximal inhibitory concentration (IC 50 ) <5g/mL): AB-1987, IC 50 =0.097g/mL; AB-2126, IC 50 =0.107g/mL; AB-3467, IC 50 =0.247g/mL; and AB-2445, IC 50 =0.753g/mL (Fig6A). All of the neutralizing antibodies were more potent than S309 (IC 50 =1.2g/mL) and much more potent than EY6A (IC 50 >10g/mL). In a Vero E6-based neutralization assay the same 4 antibodies potently neutralized live SARS-CoV-2 virus: AB-1987, IC 50 =0.054g/mL; AB-2126, IC 50 =0.106g/mL; AB-3467, IC 50 =0.179g/mL; and AB-2445, IC 50 =0.370g/mL (Fig6B).

Again, all 4 neutralized much more potently than EY6A (IC 50 >10g/mL), and CR3022 had no measurable SARS-CoV-2 neutralizing activity. Notably, of the 14 CoV2/RaTG13/CoV1 broadly cross-reactive antibodies that competed with EY6A, only AB-3467 neutralized SARS-CoV-2 virus (Fig5A,B) . Thus, the most consistently used IGHV/IGKV pairs in class 4 antibodies encoded antibodies that bound strongly but did not neutralize.

The affinities of selected antibodies to the different sarbecovirus RBDs were measured using biolayer interferometry. AB-3467 strongly bound CoV2 (K D =4.nM), pangolin, mink, This highlighted that class 4 antibodies were frequently non-neutralizing and that neutralizing potency was associated with ability to interfere with ACE2 binding.

To understand how AB-3467 simultaneously competed with ACE2 and with CR3022/EY6A, we solved the crystal structure of the AB-3467 Fab bound to CoV2 RBD (Fig6E,F). The structure revealed that >90% of the RBD binding interface was provided by the antibody heavy chain, which projected a long CDRH3 loop containing three tyrosine (Y) residues and a tryptophan (W) residue whose side chains contacted hydrophobic components of the RBD surface (Fig6G). Much of this surface (~80%) was also contacted by the heavy chain of CR3022 (Fig6E and Table S3B ). However, unlike CR3022, the backbone of AB-3467's long CDRH3 loop extended the antiparallel beta-sheet running through the centre of the RBD J o u r n a l P r e -p r o o f domain (Fig6G). The sole light chain contribution was mediated by side chain interactions of CDRL1 residue Y32 with the highly conserved RBD residue R408.

As a consequence of CDRH3 adding an antiparallel beta sheet to the RBD core, the spatial orientation of the AB-3467 heavy and light chains with respect to the RBD surface was rotated ~180 degrees relative to that of CR3022 (Fig6E). Consequently, the largely noncontacting AB-3467 light chain was projected towards the ACE2 epitope (rather than away from it), thus facilitating blockage of ACE2 binding due to steric hindrance. Inspecting deposited datasets, this rotated orientation has been seen in five other class 4 antibodies, all of which show potent SARS-CoV-2 neutralization likely through similar ACE2 steric hindrance by the light chain and each employing a long CDRH3 loop (Jette et al., 2021; Liu et al., 2020; Saunders et al., 2021; Wrapp et al., 2020) . (Jette et al., 2021) . As such, antiparallel beta-sheet extension at the RBD surface, mediated through long heavy chain CDR3s, appeared to be a common mechanism of potent neutralizing class 4 antibodies. In contrast, shorter CDRH3s were present in all 14 of the class 4 antibodies with recurrent IGHV/IGKV pairs that we had expressed and confirmed broad sarbecovirus RBD reactivity but no neutralizing activity (Fig7A). The CDRH3 length of AB-3467 was also considerably longer than the mean length of the total sequenced human or mouse cross-reactive sequences (FigS6H,I).

As noted above, AB-3467 was one of 260 CoV1/CoV2 RBD-binding and sequenced B cells comprising a large, highly mutated clone induced by repeated CoV1 RBD-immunization (Fig5C). The observed VH4-59/VK1-9 pairing was infrequent in the RBD response of other mice (TableS2). VH4-59 paired with VK1-9 accounted for 0.5% of the total naïve repertoire in Ig-humanized mice, however only a minority of naïve B cells had CDRH3 as long as AB-3467 (Fig7B), including naïve B cells expressing the VH4-59 heavy-chain (FigS6K). To further explore the VH4-59 VK1-9 antibodies, an additional 17 antibodies were expressed from the same clonal lineage as AB-3467, as well as the unmutated common ancestor (Fig5C). All 17 of these antibodies and the unmutated common ancestor bound to CoV2, RaTG13, CoV1 and mink RBDs and blocked EY6A, confirming the diverse patterns of somatic mutations preserved a broad reactivity to the conserved class 4 epitope that had been J o u r n a l P r e -p r o o f established during VDJ-recombination (TableS3B). The unmutated ancestor nevertheless had 100-fold lower affinity for CoV2 RBD (K D =499nM) than its hypermutated descendants AB-3467 (K D =4.1nM) and AB-3623 (K D =2.3nM) (FigS5). In addition, we expressed five clonally unrelated VH4-59 VK1-9 antibodies with different D and J segments and shorter CDRH3s but matched CDRH1/2s and CDRL1/2s. These came from B cells sorted from CoV1 RBD-immunized mice on days 7 or 20 that did not cross-react with CoV2 RBD tetramers and were much less effective at cross-reactive binding or blocking EY6A than the unmutated ancestor of AB-3467 (TableS3B). Compared to AB-3467, the highest spike binder Class 4 antibodies neutralized SARS-CoV-2 variants-of-concern.

The binding footprint of AB-3467 was highly conserved across 192,000 sarbecovirus RBD genomes (TableS1A), and separate from the mutations on the rim of the RBS, including K417N, E484K or N501Y, that provide an escape from class 1 and class 2 antibody binding (Fig6F). In biolayer interferometry, the affinity of AB-3467 was neither hindered by K417N, (Fig7E), with a maximum of a two-fold drop in IC 50 against any of these variants. By contrast, class 1, 2 and class 3 antibodies show 10-fold Greaney et al., 2021) or 4-6 fold drops in potency to these lineages.

The importance of vaccines that target conserved epitopes has become apparent with the increasing predominance of viral mutations that decrease the neutralizing efficiency of antibodies against SARS-CoV-2, particularly towards epitopes around the RBS Liu et al., 2021) . Given the ongoing risk of SARS-CoV-2 evolution in animal hosts (Oude Munnink et al., 2021) and potential for emergence of related viruses (Wahl et al., 2021; Wang et al., 2018) , elicited antibodies should target not only the current SARS-CoV-2 lineages but display resistance to novel variants of concern.

Of the cross-neutralizing epitopes on RBD, antibodies targeting the class 4 epitope (Barnes et al., 2020) hold considerable hope at fulfilling this goal. This epitope is highly conserved, possibly because of its interaction with the S2 subunit in the pre-fusion spike conformation when the RBD is in the "down" state. Antibodies to this epitope show considerable affinity for both SARS-CoV-1 and SARS-CoV-2 and the bat precursor virus RaTG13 , highlighting their potential resistance to future mutational escape.

Here, we showed that immunization strategies with divergent sarbecovirus RBDs induced antibodies to the conserved class 4 epitope with a range of recurring IGHV/IGKV combinations. These antibodies were not only stimulated by CoV2 RBD immunizations, but also by immunization with diverse sarbecovirus RBDs, extending recent work showing nanoparticles containing WIV1, Rf1, RmYN02 and Pang17 RBDs could result in CoV2 specific antibodies . In this study we showed that immunization CoV1 and RaTG13 RBD actually resulted in a higher proportion of the response targeting this class 4 epitope compared with immunization with CoV2 RBD. The potently neutralizing class 4 AB-3467 antibody lineage described here, similar to previously described cross-reactive antibodies (Pinto et al., 2020; Rappazzo et al., 2021; ter Meulen et al., 2006; Wec et al., 2020; Wrapp et al., 2020) , was induced by CoV1 RBD exposure. Optimal RBD immunization strategies could therefore require vaccinations with non-CoV2 RBDs. Our results addressed the nature of CoV1 RBD-cross-neutralizing polyclonal antibodies detected in the sera of macaques immunized with CoV2 RBD conjugated to ferritin in protein nanoparticles or with isolated CoV2 RBD given as mRNA vaccine (Saunders et al., 2021) . The serum titers are 20fold lower than against CoV2, indicating a minority of the response recognizes conserved epitopes, but these antisera did measurably block binding of a class 4 antibody DH1047 more than convalescent patient serum. Later manuscripts have emerged suggesting the macaque J o u r n a l P r e -p r o o f repertoire to be pre-disposed towards the generation of CoV1/CoV2 cross-reactive responses compared to humans and mice, and suggesting caution interpreting macaque models of crossreactive nAbs compared to data elicited from human antibody repertoire mice (He et al., 2021) . In contrast, we found the IGHV regions accounting for 76% of class 4 antibodies were equally represented in the naïve repertoire of humans and Ig-humanized mice, suggesting these as a more appropriate model of cross-reactive responses. While these previous studies provide limited analysis of the nature of class 4 antibodies elicited by RBD immunization, and are limited to CoV2 RBD immunization, their results complement the findings here that second generation immunization strategies employing diverse RBD's elicit a small but consistent expansion of class 4 antibodies with long heavy chain CDR3's targeting the RBD beta sheet backbone to confer broad reactivity and orient light chains for steric blockade of the ACE2 binding site.

Although we were able to identify and express numerous antibodies to the class 4 epitope, most bearing high affinity to diverse sarbecovirus RBDs, the great majority were not neutralizing. This corresponds with the characteristics of antibodies isolated from convalescent patients in the published antibody database (CoV-AbDab) (Raybould et al., 2021), wherein 17% (428/2523) of antibodies described to bind CoV2 also bind to CoV1 but only 6% (57/899) of neutralizing SARS-CoV-2 antibodies show any binding to CoV1. Taken together, these results highlight the challenges to generating neutralizing antibodies to conserved sarbecovirus epitopes. AB-3467 and four neutralizing class 4 antibodies that have been isolated from convalescent humans reveal a shared mechanism: concurrently binding the class 4 epitope and blocking ACE2 interactions with the RBS by steric hindrance (Jette et al., 2021; Liu et al., 2020; Saunders et al., 2021) . In this study we found and characterized two antibodies induced by immunization with these unusual properties and similarly broad and potent neutralizing activity, AB-3467 and AB-3623 from the same expanded clone but with highly divergent CDR1 and CDR2 mutational profiles. By contrast antibodies with the same IGHV and IGKV but different CDRH3 sequences lacked these properties. The other described neutralizing class 4 antibodies show a wide range of IGHV/IGKV/IGLV pairings (Jette et al., 2021; Liu et al., 2020; Wrapp et al., 2020) indicating this unusual mechanism for virus neutralization can be conveyed by a wide variety of V-regions and mutational profiles.

Despite their divergent CDR1 and CDR2 sequences, AB-3467 and the other potently neutralizing class 4 antibodies so far described possess an extended CDRH3 which forms a J o u r n a l P r e -p r o o f fifth antiparallel beta-sheet extending the four antiparallel beta sheets at the core of the coronavirus RBD. Antiparallel binding of CDRH3 orientates the antibody H-chain so that the light chain is projected towards the ACE2 receptor binding site, whereupon it sterically blocks ACE2 binding without depending on contacts with the variable class 1 and class 2 antibody epitopes. AB-3467 was distinct in this manner because its binding interface consisted almost exclusively of the antibody heavy chain. As such AB-3467 was able to function similarly to camelid single domain antibodies (nanobodies) , while still maintaining the advantages of being a fully human antibody.

A focus of next generation vaccination strategies should be the induction of class 4 antibodies that prevent ACE2 interactions through steric hindrance, similarly to the AB-3467 clonal lineage induced here. The focus on highly conserved spike protein structural elements provides these antibodies with resistance to viral mutations, superior to that of antibodies binding the class 1, 2 and class 3 epitopes which show significant reductions in potency to emerging strains Greaney et al., 2021) . However, the requirement for a long CDRH3 makes these antibodies less common in the circulating naïve B cell repertoire.

Indeed, our flow cytometric analyses indicated that a small percentage of heterologous RBDelicited class 4 antibodies block ACE2 binding, and we found only a single expanded clone of 260 cells with these properties out of >7500 sequenced RBD-binding GC B cells, in contrast to non-neutralizing class 4 antibodies which were frequently induced by a range of immunization regimes and employed more common CDRH3 lengths and IGHV/IGKV pairs. Long CDRH3s increase the likelihood of self-reactivity (Mouquet et al., 2010) and downregulation of surface IgM on naïve B cells , so that uncommon B cell precursors of neutralizing class 4 antibodies may need especially potent stimulation to overcome tolerance checkpoints and undergo clonal redemption from self-reactivity by somatic hypermutation (Burnett et al., 2018) . The findings here in humanized mice set out a structurally defined pathway for these successful next-generation COVID immunization strategies.

Sheep erythrocyte immunizations represent a well characterized, adjuvant free method for eliciting potent, reproducible antibody responses in rodents (Paus et al., 2006; Yi et al., 2015) , and have been shown to elicit potent antibody responses in humans (Leikola and Aho, 1969) . However, the antigenic complexity of xenogeneic erythrocytes and their cross-J o u r n a l P r e -p r o o f reactivity with human blood cell antigens may preclude their use for large scale vaccination (Hoffman et al., 1973; Villa and De Biasi, 1983; Yi et al., 2015) . Given the evidence that CoV2 RBD mRNA immunization of macaques elicits comparable titers of CoV1-neutralizing serum antibodies to stabilized spike mRNA vaccination (Saunders et al., 2021) , the class 4 antibody responses demonstrated here may be more potently elicited by second generation mRNA vaccines encoding CoV1 RBDs tethered to the plasma membrane. See also Figure S6 J o u r n a l P r e -p r o o f

Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Deborah L. Burnett (d.burnett@garvan.org.au) .

CR3022 knock in mice generated in this study are available upon request and completion of a routine MTA.

Crystallography for this project is uploaded to the PDB. Accession numbers are listed in the KRT.

The published article includes all datasets generated or analyzed during this study, including full SC-RNAseq data in the accompanying tables, figures and supplementary material. (Yang et al., 2014) . Embryos for microinjection were produced by mating stud males with super-ovulated C57BL/6J females. Stud males were wild-type C57BL/6J mice. All embryos were microinjected with in vitro transcribed and polyadenylated mRNA encoding S.pyogenes Cas9 and either one or two in vitro transcribed sgRNAs. Microinjected embryos were cultured overnight and introduced into pseudo-pregnant foster mothers. Pups were screened by PCR and Sanger sequencing of ear-punch DNA to identify founder mice which were then crossed to C57BL/6J mice to establish each line.

CoV sequences were taken from the literature in the field (Amanat et al., 2020; Zhang et al., 2020) . Open reading frame of CoV proteins were cloned into the pCEP4 mammalian expression vector (Thermo Fisher Scientific) with a N-terminal IgG leader sequence and C- (Amanat et al., 2020) . The protein was further purified on a Sepharose 6 gel filtration column (GE Healthcare) using an AKTA Pure FPLC instrument (GE Healthcare) to isolate the trimeric protein.

Proteins were biotinylated by incubating for 30 min at room temperature with EZ-Link NHS-PEG4-Biotinylation reagent (Thermo Fisher Scientific) at a 10:1 biotin-to-protein ratio. Free biotin was removed from the samples by repeating the buffer exchange step in a second ZebaSpin column equilibrated with PBS. Following biotinylation proteins were conjugated to the relevant streptavidin fluorophores at a 4:1 molar ratio for 1h at 4 degrees. Excess biotinylated protein that was not coupled to streptavidin was removed by size exclusion through 30KMWCO Amicon Ultracel centrifugal columns (Merck Millipore Ltd). All recombinant protein fluorophores were titrated prior to use. Tetramers were stored for a maximum of 4 weeks prior to use.

Antibodies were transiently expressed as human IgG1 in HEK293 cells using standard plate transfection and the Expi system (LifeTechnologies) and purified with protein G Sepharose (Genscript) according to the manufacturers' recommendations. Buffer exchange was performed using Genescript desalting preparation columns. Quality control of proteins included SDS-Page and western blots. Samples that failed to meet QC or were unable to be expressed at high concentrations (<0.01mg/mL) were removed from the analysis. Samples were frozen at -80 degrees prior to use.

Purified monoclonal IgG antibodies were buffer exchanged into PBS using equilibrated (Kabsch, 2010) . Space groups were determined with POINTLESS (Evans, 2011) and scaling and merging were performed with AIMLESS (Evans and Murshudov, 2013) , both part of the CCP4 suit of software . Data collection statistics are shown in Table S4 . Structures were determined by molecular replacement using PHASER (McCoy et al., 2007) . The search model for the RBD component was derived from PDB entry 7kzb (Rouet et al., 2021) , whilst the Fab components were derived from PDB entry 7czx where the heavy and light chains were split into variable (VH + VL) and constant (CH1 + CL) domain pairings. Two RBD-Fab complexes were found in the asymmetric unit, consistent with a solvent content of ~54%. The model was iteratively improved via rounds of refinement performed with REFMAC5 (Murshudov et al., 2011) and manual real-space inspection and adjustment performed with COOT (Emsley et al., 2010) . Model and refinement statistics are shown in Table S4 . The final model comprises two essentially identical RBD-Fab complexes.

Conjugation was performed as described previously (Burnett et al., 2018) . For assessment of the effects of EDCI on RBD proteins CoV2 RBD was resuspended at 10ug/mL in 1mL conjugation buffer. 10mg of EDCI was added (or sample left as a control) and the sample was mixed for a further 30 minutes on ice. Following this incubation 2ug/mL EDCI conjugated or unconjugated CoV2 RBD was added to 1 million splenocytes from a mouse with a knocked in CR3022 heavy and light chain B cell receptor and sample was incubated for a further 30 minutes. Samples were then washed by the addition of 100uL PBS and centrifuged at 2,300 rpm (1,111 g) for 1 min at 4°C. Samples were then incubated with S309-AF647 or ACE2FC-BV605 generated as described above for 30 minutes, followed by a further wash step and transfer to flow tubes.

On the day of harvest organs were collected into PBS with 1% Bovine Serum Albumin Biolegend) . Live dead discrimination was performed with 7AAD (Biolegend). B cells were determined as TCRB -, CD4 -, CD11b -, B220 + cells.

Germinal centers were identified as Fas + , CD38 -, IgD -.

Cells were filtered using 35µm filter round-bottom FACS tubes (BD Pharmingen)

immediately before data acquisition on a LSR II analyzer (BD Pharmingen) sorted samples were analyzed on a FACS ARIA II or III (BD Pharmingen). Forward-and side-scatter threshold gates were applied to remove red blood cells and debris and approximately 2-5 × 10 6 events were collected per sample. Cytometer files were analyzed with FlowJo software (FlowJo LLC, Ashland, Oregon, USA).

For flow cytometric epitope binding assays spleens from immunized day 20 TRIANNI mice were incubated with SARS-CoV-2 RBD at 200ng/mL in 1% BSA. Unimmunized mice with knocked in CR3022 heavy and light chain B cell receptor were used as a staining control as affinity maturation of the CR3002 antibody has been shown to be able to mature to bind different epitopes (Rouet et al., 2021) . IgG1-BUV395 (10.9, BD Biosciences), Fas-PeCy7 Biolegend) as well as S309 conjugated to AF647 and EY6A conjugated to AF488. Live dead discrimination was performed with 7AAD (Biolegend). Samples were then incubated with ACE2FC-BV605 generated as described above for 30 minutes. Between stains samples were washed by the addition of 100uL PBS and centrifuged at 2,300 rpm (1,111 g) for 1 min at 4°C. Gating of germinal center and memory B cells was performed as above.

For analysis of human samples, cryopreserved PBMCs were thawed rapidly in a 37-degree waterbath and washed with pre-warmed RPMI media supplemented with 2 mM L-glutamine, IU/mL penicillin, 50μg/mL streptomycin and 10% heat inactivated fetal calf serum (Sigma).

The cells were resuspended in PBS and counted. Single cell suspensions were labelled with the following anti-human antibodies at the indicated dilutions 1:10 CD21-BV421 (B-ly4, BD Bioscience), 1:10 IgD-BV510 (IA6-2, BD Bioscience), 1:10 CD10-BV605 (HI10a, BD Bioscience), 1:20 CD19-BV711 (SJ25C1, BD Bioscience), 1:10 CD20-APC-H7 (H27, BD and sequenced using a 150-cycle High-Output Kit. To process the sequencing data, we used the 10x Genomics cellranger pipeline (v2.1.0), comprising the mkfastq stage.

Using cellranger mkfastq, raw base call files were demultiplexed into sample-specific FASTQ files. FASTQs were then processed with 10x Genomics cellranger vdj (v4.0.0) using a custom reference that included the human IGH and IGK variable (V/D/J) genes and the mouse Igh and Igk constant region genes. The resulting VDJ contigs were post-processed using stand-alone IgBLAST (v1.14) (Ye et al., 2013) to generate additional alignment details.

Clonal lineages were defined by independent clustering of the IGH and IGK using cd-hit (Li and Godzik, 2006) . IGH and IGK were subset by V gene, J gene and CDR3 length and CDR3 nucleotide sequences were clustered at 90% identity. Clonal trees were generated using the linearham package (Dhar et al., 2020) . Trees were visualised using the ggtree package (Yu, 2020) .

Following flow cytometry sorting of single cells into 96-well plates, Smart-Seq2 was performed by following the protocol of Picelli et al. (Picelli et al., 2014) with the following modifications. Reactions were performed at half volumes, the IS PCR primer was reduced to J o u r n a l P r e -p r o o f 50nM final concentration and the number of PCR cycles increased to 28. Sequencing libraries were prepared using the Nextera XT Library Preparation Kit (Illumina) at one quarter of the recommended volume. Sequencing was performed using the Illumina NextSeq 500 instrumet with 150 bp paired-end reads to a median depth of ~1 million reads per cell.

Paired heavy and light chain sequences were assembled with mixcr. Paired fastqs for each cell were processed using mixcr (v3.0.9) using the analyze command with the shotgun option (Bolotin et al., 2017) . The resulting contigs were then processed with IgBLAST as for the 10x VDJ datasets. Analysis of IGHV gene usage IGHV gene usage, as defined by IgBLAST, was compared to relevant naïve repertoires. For humans, IGH repertoire sequencing from 114 healthy human controls was obtained from SRA (BioProject: PRJNA491287) and processed as previously reported (Nielsen et al., 2020) . The naïve compartment was defined as unmutated (<0.5% median SHM) IgM clones and donors with at least 1000 unique IgM clones were retained (n=100). For TRIANNI mice, naïve B cells were sorted from an unimmunised mouse and sequences and analysed using the 10x Genomics platform. To test if an IGHV gene's usage differed between a response and the underlying naïve repertoire, the odds ratio and P-value were determined by a Fisher's exact test using the fisher.test function from the stats package in R and plotted using the ggplot package with RStudio.

The IGHV usage among patient derived mAbs that bind RBD were extracted from the CoV-AbDab (release 16 th June 2021) (Raybould et al., 2021) . The csv file was downloaded and subset for mAbs meeting the following criteria: Binds to = SARS-CoV2 or SARS-CoV1, Protein + Epitope = RBD, Heavy V Gene = human IGHV, Origin = B-cells from COV2

Patient. IGHV gene usage was quantified using the reported 'Heavy V Gene' and neutralization information was collected from the 'Neutralizing Vs' field. Data manipulation was performed in R using the tidyverse package.

Hemizygous male and female K18-hACE2 mice (B6.Cg-Tg(K18-hACE2)2Prlmn/J, JAX stock #034860) were obtained from Jackson Laboratory. Mice were housed in groups and fed normal rodent chow. At 6-8 weeks age, mice were intranasally inoculated with 1 x 10 4 PFU SARS-CoV-2 (Isolate AUS/VIC01/2020) in a 30 µL volume. Mice were weighed and monitored twice daily. Once mice lost 20% body weight, or showed any severe clinical disease, they were humanely euthanized and tissues collected for downstream processing.

For SARS-CoV-2 viral-cell fusion assays, stable ACE2-expressing Hek293T cells were generated by lentiviral transductions and lentiviral particles pseudotyped with SARS-CoV-2 Spike envelope were produced by co-transfection with a GFP encoding lentiviral plasmids.

Neutralization activity of sera was measured using a single round infection of ACE2-HEK293T with Spike-pseudotyped lentiviral particles. Virus particles were incubated with serially diluted antibodies for 1 hour at 37°C and then added onto ACE2-HEK293T cells.

Following spinoculation at 1200g for 1 hour at 18°C, the cells were moved to 37°C for 72 hours. Entry of Spike particles was imaged by GFP-positive cells (InCell Analyzer) followed by enumeration with InCarta software (Cytiva, USA). Neutralization was measured by J o u r n a l P r e -p r o o f reduction in GFP expression relative to control group infected with the virus particles without antibody treatment.

Antibodies were serially diluted and mixed in duplicate with an equal volume of virus solution at 1.25x10 4 TCID50/mL. After 1 hour of virus-antibody coincubation at 37°C, 40μL

were added to an equal volume of Vero E6 or HEK293T cells (5x10 3 cells in suspension) in 384-well plates for a final MOI=0.05. After 72h, cells were stained with NucBlue (Invitrogen, USA) and the entire well was imaged with InCell Analyzer microscopy system (Cytiva). Nuclei counts were obtained for each well with InCarta software (Cytiva) as a proxy for measuring cytopathic effect. Counts were compared between test antibody, mock controls (defined as 100% neutralization), and infected controls (defined as 0% neutralization). Sample-mediated neutralization was calculated using the formula; % viral neutralization = (D-(1-Q))x100/D, where Q = nuclei count normalized to average of mock controls, and D = 1-Q for average of infection controls.

GraphPad Prism 8 (GraphPad Software, San Diego, USA) was used for data analysis. When the data were normally distributed, an unpaired Student's t-test was performed for analysis.

When data was not normally distributed Welsh's correction was applied. For all tests, P < 0.05 was considered as being statistically significant. Unless otherwise stated error bars represent arithmetic mean. For all figures, data points indicate individual mice. * represents P < 0.05, ** represents P < 0.01, *** represents P < 0.001, **** represents P < 0.0001.

J o u r n a l P r e -p r o o f

Figs. S1 to S6 Figure S1 . Validation of experimental model (refers to figure 1). Figure S2 . B cell response elicited by RBD-focused immunization included broadly crossreactive cells in standard inbred lines (refers to figure 1 ). Figure S3 . B cell response elicited by RBD-focused immunization included broadly crossreactive cells in Ig-humanized mice (refers to figure 1 ). Figure S4 . B cell response to coronavirus RBDs in Ig-humanized mice included recurrent IGHV/IGKV pairs (refers to figure 3 and figure 4) . Tables S1 to S4 Table S1 . Sequence identity and binding residues of neutralizing class 4 antibodies (refers to figures 3 and 5). Table S3 . Binding data of expressed antibodies (refers to figure 5). Table S4 . X-Ray Diffraction Data Collection Statistics (refers to figure 6 ).

J o u r n a l P r e -p r o o f

IMGT((R)) tools for the nucleotide analysis of immunoglobulin (IG) and T cell receptor (TR) V-(D)-J repertoires, polymorphisms, and IG mutations: IMGT/V-QUEST and IMGT/HighV-QUEST for NGS

A serological assay to detect SARS-CoV-2 seroconversion in humans

Immunodominant antibody germlines in COVID-19

Antibody repertoire analysis of mouse immunization protocols using microfluidics and molecular genomics

Zooanthroponotic potential of SARS-CoV-2 and implications of reintroduction into human populations

SARS-CoV-2 neutralizing antibody structures inform therapeutic strategies

Antigen receptor repertoire profiling from RNA-seq data

Evolutionary origins of the SARS-CoV-2 sarbecovirus lineage responsible for the COVID-19 pandemic

Germinal center antibody mutation trajectories are determined by rapid self/foreign discrimination

Resistance of SARS-CoV-2 variants to neutralization by monoclonal and serum-derived polyclonal antibodies

Mosaic nanoparticles elicit cross-reactive immune responses to zoonotic coronaviruses in mice

The mouse antibody heavy chain repertoire is germline-focused and highly variable between inbred strains

A Bayesian phylogenetic hidden Markov model for B cell receptor sequence analysis

Features and development of Coot

An introduction to data reduction: space-group determination, scaling and intensity statistics

How good are my data and what is the resolution?

Multiple SARS-CoV-2 variants escape neutralization by vaccine-induced humoral immunity

Complete Mapping of Mutations to the SARS-CoV-2 Spike Receptor-Binding Domain that Escape Antibody Recognition

Change-O: a toolkit for analyzing large-scale B cell immunoglobulin repertoire sequencing data

Vaccine Breakthrough Infections with SARS-CoV-2 Variants

Broadly neutralizing antibodies to SARS-related viruses can be readily induced in rhesus macaques. bioRxiv

Production of antibody to sheep red blood cells by human tonsil cells in vitro

Neutralizing activity of Sputnik V vaccine sera against SARS-CoV-2 variants

Broad cross-reactivity across sarbecoviruses exhibited by a subset of COVID-19 donor-derived neutralizing antibodies

Animal and translational models of SARS-CoV-2 infection and COVID-19

openPrimeR for multiplex amplification of highly diverse templates

Experimentally induced mononucleosis-like heterophile antibodies in man

Cd-hit: a fast program for clustering and comparing large sets of protein or nucleotide sequences

Antibody to a Functionally Conserved Site Is Mediated by Avidity

Neutralizing Activity of BNT162b2-Elicited Serum -Preliminary Report

Covid-19 Vaccine against the B.1.351 Variant

Phaser crystallographic software

ImmunoGlobulin galaxy (IGGalaxy) for simple determination and quantitation of immunoglobulin heavy chain rearrangements from NGS

Polyreactivity increases the apparent affinity of anti-HIV antibodies by heteroligation

REFMAC5 for the refinement of macromolecular crystal structures

Human B Cell Clonal Expansion and Convergent Antibody Responses to SARS-CoV-2

Transmission of SARS-CoV-2 on mink farms between humans and mink and back to humans

Antigen recognition strength regulates the choice between extrafollicular plasma cell and germinal B cell differentiation

A pair of noncompeting neutralizing human monoclonal antibodies protecting from disease in a SARS-CoV-2 infection model

Full-length RNA-seq from single cells using Smart-seq2

Cross-neutralization of SARS-CoV-2 by a human monoclonal SARS-CoV antibody

Broad and potent activity against SARS-like viruses by an engineered human monoclonal antibody

Potent SARS-CoV-2 binding and neutralization through maturation of iconic SARS-CoV-1 antibodies

Neutralizing antibody vaccine for pandemic and pre-emergent coronaviruses

Human monoclonal antibody combination against SARS coronavirus: synergy and coverage of escape mutants

Antigen-dependent colonies of human peripheral blood lymphocytes: an immunomorphologic study

SARS-CoV-2 infection is effectively treated and prevented by EIDD-2801

Serological Evidence of Bat SARS-Related Coronavirus Infection in Humans

mRNA vaccine-elicited antibodies to SARS-CoV-2 and circulating variants

Broad neutralization of SARSrelated viruses by human monoclonal antibodies

Overview of the CCP4 suite and current developments

Structural Basis for Potent Neutralization of Betacoronaviruses by Single-Domain Camelid Antibodies

Serum Neutralizing Activity Elicited by mRNA-1273 Vaccine

Generating genetically modified mice using CRISPR/Cas-mediated genome engineering

IgBLAST: an immunoglobulin variable domain sequence analysis tool

Splenic Dendritic Cells Survey Red Blood Cells for Missing Self-CD47 to Trigger Adaptive Immune Responses

Using ggtree to Visualize Data on Tree-Like Structures

Structural basis of a shared antibody response to SARS-CoV-2

Recognition of the SARS-CoV-2 receptor binding domain by neutralizing antibodies

Probable Pangolin Origin of SARS-CoV-2 Associated with the COVID-19 Outbreak

Human transitional and IgM(low) mature naive B cells preserve permissive B-cell receptors

Structural basis for the neutralization of SARS-CoV-2 by an antibody from a convalescent patient

We thank the Garvan Institute Australian BioResources, Garvan Molecular Genetics, Flow cytometry and Garvan-Weizmann Single Cell facilities. We thank the Krammer lab for the plasmid encoding CoV2 Spike. Part of this work was performed on the MX2 beamline at the Australian Synchrotron, part of ANSTO. We thank L. Burnett for his comments on the manuscript. The authors would like to thank the study participants for their contribution to the research. They would like to acknowledge members of the study group: protocol steering committee: Andrew R. Lloyd (The Kirby Institute), John Kaldor